The modern technology trend is towards higher density and lower profile electrical devices. This trend has driven a demand for improving the power density of power supplies. As a result, various techniques have been developed to increase the switching frequencies of the power supply in order to reduce the size and bulk of magnetic components and filtering elements. Problems experienced with increasing the operating frequencies include higher switching losses and the generation of worse electromagnetic interference (EMI). Since the switching losses in power semiconductors are directly proportional to the operating frequency, thermal management is also a big challenge, since the space saved by using smaller filtering components is more than offset by the need for larger heat sinks.
The development of a method of soft switching power supplies addressed most of the above problems. For the DC/DC conversion stage, the soft, zero voltage switching (ZVS) method for the power switches eliminated turn on losses. At the same time, this method improved EMI performance by lowering fast rising switching currents. Thus, the method significantly improved the efficiency of the power converter and enabled switching at higher frequencies. The demand for higher power density, however, is increasing unabated. New problems are surfacing as the soft, ZVS converter is being switched at ever higher frequencies. Conventional soft switching converters switching at high frequencies are exhibiting high power losses during light load conditions. As a result, although such converters are very efficient at full load, they are prone to failure at light loads. Many semiconductor manufacturers attribute this failure to various semiconductor phenomena, such as the reverse recovery speed of the MOSFET body diode, and the construction of the MOSFET's channel, etc. One other significant problem, however, is that it is difficult to charge and discharge the output capacitance of the MOSFET bridge switching elements at light loads when the converter is in the hard switching mode. This is because the energy stored in the resonant choke in such converters is very low and therefore cannot charge/discharge the output capacitance of the bridge switching elements. This is true for converters having a primary side resonant choke or a saturable choke on the secondary side. This problem exists in all known ZVS control techniques. In fact, this drawback at light load for conventional soft switching converters is even worse than for conventional hard switching bridge converters.
FIG. 1 illustrates a worst case scenario when the power supply is at no-load. At no-load, the only energy available for ZVS switching is the magnetizing current of the power transformer. Since this current is typically very low, it cannot charge/discharge the MOSFET switch capacitance in the required delay time.
As shown in FIG. 1, the circuit on the left side shows a bridge circuit 10 with a transformer primary winding. Since the transformer magnetizing current is very low at a no-load condition, negligible energy is stored in the primary winding or in any series resonant choke. The series resonant choke for the circuit in FIG. 1 could be the leakage inductance of the transformer or any external inductance. Since the energy stored in the magnetizing inductance is very low, it cannot enable ZVS of the bridge elements, and thus, can be neglected in the analysis. The bridge circuit 20 on the right side in FIG. 1 shows this worst case scenario where the transformer winding has no effect.
To analyze the circuit in FIG. 1, it is first assumed that an active diagonal is operating, e.g., switches Qa and Qd are on. At the end of the active period, switch Qd will turn off. The voltage across it will not rise since there is no charging current available. As a result, the voltage across Qd will remain close to zero during the dead time at the end of the active period. Thus, the MOSFET switch Qd output capacitance, shown as capacitor Cd, is at zero volts and fully discharged and the capacitance across switch Qc, shown as capacitor Cc, is fully charged at the Bulk + voltage shown. In typically operating soft switching converters, switch Qc will be turned on after a short delay. When switch Qc turns on, the energy stored in capacitor Cc is fully discharged in switch Qc. At the same time, as the lower end of the switch Qc rises to the Bulk + voltage level, the capacitance Cd of the lower switch Qd also gets charged through switch Qc. Thus, there are two kinds of resistive losses in switch Qc: one due to the discharging of capacitor Cc and other due to the charging of capacitor Cd. These losses result in power dissipation which is proportional to the operating frequency as represented in the following formula:Pturn-on=(0.5×Cc×(Vbulk)2×Fsw)+(0.5×Cd×(Vbulk)2×Fsw)Where Fsw is the switching frequency. Assuming Ca=Cb=Cc=Cd:Pturn-on=Cc×(Vbulk)2×Fsw
These resistive losses, and the resulting power dissipation, may be tolerable at lower switching frequencies in the range of 100 kHz to 200 kHz. At much higher frequencies, e.g., above 400 kHz, however, these losses predominant such that the total power lost in the bridge switches at light loads exceeds the losses at full-load. This predominance is illustrated in FIG. 1A for a typical soft switching full bridge converter. FIG. 1A shows total losses in the entire converter versus the load percentage. At light loads, most of this total loss is due to losses in the bridge switching devices.
FIG. 2 is a circuit diagram of an exemplary prior art full bridge power converter 30 where a primary side resonant inductor is used for achieving soft, zero voltage switching. As is seen, a resonant inductor Lr is inserted in series with the primary of the power transformer. Inductor Lr could also be the parasitic leakage inductance of the transformer. A simplified representation of switches Qa, Qb, Qc, and Qd is shown in FIGS. 2–4 such that the switch capacitances of the corresponding switches are not shown. The existence of the switch capacitances is well known in the art. For reference, the switch capacitances are as shown in bridge 20 for switches Qa, Qb, Qc, and Qd in FIG. 1. FIG. 2A is a set of voltage and current waveforms illustrating the operation of the power converter in FIG. 2.
During the active period of the switching diagonal, e.g., Qa and Qd are on, energy is stored in inductor Lr due to the primary current flowing through it. When one of the diagonal MOSFET switches (e.g., Qd) turns off, the energy stored in inductor Lr is used to charge that MOSFET's output capacitance and to discharge the output capacitance of the other MOSFET in the same vertical leg. As a result, ZVS action is achieved.
In addition to the fact that the circuit topology in FIG. 2 has the drawback of losses at light load at higher frequencies, since the capacitance of each MOSFET switch is intrinsic and does not change with frequency, the size of the resonant inductor Lr is independent of frequency. As a result it may be quite large for a high frequency power supply. Inductor Lr is also lossy since it handles very high primary full load currents and its flux swings in both directions, generating high core losses. The series inductor in FIG. 2 also introduces a delay, e.g., 200 nS, which reduces the available maximum duty cycle of the converter. This delay is a serious drawback at higher frequencies.
FIG. 3 is an exemplary prior art full bridge converter 40 where two saturable inductors, Ls1 and Ls2, are connected in series with the secondary side's rectifier diodes. In operation, for an active transformer period when diagonal Qa-Qd is conducting, the dotted end of the secondary is positive and D31 is forward biased, providing current to the output load through inductors Ls1 and Lout. This current saturates inductor Ls1. At the end of the active period, Qd turns off and the secondary voltage starts to fall. Since Ls1 is saturated and Ls2 is in blocking mode since D32 is reverse biased, this forces the current in output inductor Lout to keep flowing through the upper half of the secondary, i.e., D31-Ls1. This DC inductor current also has an AC component in the form of ripple current. The transformer action causes this ripple current to be reflected back to the primary side, which forces the primary current to keep flowing while achieving ZVS action. Similar ZVS action is repeated by Ls2 in the next active period.
Although the circuit in FIG. 3 achieves ZVS action satisfactorily at higher loads, it still has the drawback of losses at light load at higher frequencies. At frequencies above 200 kHz, for example, the core losses in the secondary side saturable cores of Ls1 and Ls2 are very high and could result in thermal runaway for the square loop amorphous cores typically used. The blocking effect of these saturable inductors also reduces the available duty cycle.
FIG. 4 shows a prior art full bridge converter 50 including two external resonant inductors Lr11, Lr21 and two split capacitors C11, C21 to generate a split bulk + voltage rail. FIG. 4A is a set of exemplary voltage and current waveforms illustrating the operation of the power converter in FIG. 4 for exemplary circuit values as shown. In operation, when the diagonal full bridge devices, e.g., Qa, Qd, are in conduction, current flows in the respective inductor (Lr11, Lr21) and energy is stored. At the end of the active period when the switch, e.g., Qd, turns off, the energy stored in the inductor is utilized to achieve the ZVS transition.
The prior art converter 50 shown in FIG. 4 may provide zero voltage switching down to very light loads for all four full bridge MOSFETS, Qa, Qb, Qc and Qd, if the power transformer is non-ideal, i.e., has high leakage inductance, and thus may be able to address the problem of losses at light load. However, this circuit has several drawbacks. The ripple current stress on the capacitors can be significant, such that capacitors of higher cost are required. Instead of using such costly capacitors, each of these bulk capacitors can alternatively be split into a series combination of two. The drawback of this solution is that this greater number of capacitors will occupy a larger volume, thereby creating an inefficient use of the available space. Another drawback of the circuit in FIG. 4 is that any inequality between the values of C11 and C21 or between the values of Lr11 and Lr21 can create problems with the current mode control of the circuit.
Another drawback of this circuit is as follows. The circuit in FIG. 4 can provide satisfactory ZVS transition from the active to the passive state. During the transition from the passive to the active state, however, the energy in inductors Lr11 and Lr21 can flow through the transformer and be transferred to the load instead of achieving ZVS transition of the passive to active leg. This drawback is lessened in applications having a large transformer leakage inductance, but for transformers with very low leakage inductance, this problem in the converter 50 shown in FIG. 4 may result in some hard switching of one leg of the bridge.
The heat sink for most power supplies is designed to accommodate heat dissipation at full-load. Although a cooling fan is typically provided for the power supply, the fan is typically controlled such that fan speed is a function of the load. Thus, at light loads, the dissipation in bridge switches is higher than at full-load and much less cooling air is available. As a result, these devices may fail due to thermal runaway. Prior art devices have addressed this failure mode through a “cold plate” approach. In this approach, the bridge switches are mounted on the same large heat sink used for cooling the boost converter or secondary rectifiers. Since the power losses in the boost converter or output rectifiers are negligible at light loads, the large heat sink can handle the extra losses in the bridge devices at light load, and thereby avoid thermal runaway. This cold plate approach is inefficient and cannot meet more demanding efficiency requirements at light load conditions. The cold plate approach also complicates the construction of the power supply as several safety requirements must be met as well, e.g., requiring insulation on the secondary side, thus rendering this approach inconsistent with high density requirements.
A circuit is therefore needed which solves the above described drawbacks of losses at light load in high frequency soft switching power converters.